Anthropic releases Claude Fable 5 model with improved benchmarks and new safety restrictions
Model release ● Confirmed 85% confidence first seen
Anthropic released Claude Fable 5, a safer version of the unreleased Claude Mythos 5 model that achieves top benchmark scores across multiple domains. The model is priced at approximately twice the cost of Claude Opus and features both transparent and hidden safety filters, including restrictions on AI research queries. It is available on subscription plans through June 22 before transitioning to pay-per-use pricing.
Decision brief
- What changed
- Anthropic released Claude Fable 5, described as a safety-restricted version of an unreleased, more capable model called Claude Mythos 5; Fable 5 tops public benchmarks, is priced at roughly twice the cost of Claude Opus, and includes both disclosed and undisclosed ('hidden') safety filters that specifically degrade performance on AI/ML research queries. It is currently included in Claude subscription plans but will move to pay-per-use pricing after June 22.
- Why it matters
- The pricing shift and restricted access to the more capable Mythos model directly affect cost forecasting and vendor evaluation for teams using Claude for complex, token-heavy work like codebase migrations or autonomous research, where the coverage says gains are concentrated. The existence of undisclosed filters that quietly lower performance on AI-research queries raises transparency and trust concerns for technical and security teams relying on benchmark claims, and the looming subscription-to-pay-per-use transition creates near-term budget uncertainty.
- Evidence
- Four independent newsletters (The Algorithmic Bridge, Interconnects, Ben's Bites, Deep Learning Weekly) report the same core facts—Fable 5 as a restricted variant of Mythos 5, top benchmark scores, 2x Opus pricing, and a June 22 pricing transition—giving reasonably consistent, though not primary-source, corroboration; two sources (Interconnects, Ben's Bites) specifically flag the hidden safety filters and community controversy.
- What remains uncertain
- It's unverified whether or when Claude Mythos 5 itself will be released, how enterprise customers' access or filtering differs from standard users, and the precise magnitude of performance degradation caused by the undisclosed filters. The claim that most non-enterprise users won't see measurable improvement in everyday tasks is asserted by only one source and not independently benchmarked here.
- Monitor next
- Watch how pricing and usage patterns shift after the June 22 transition from subscription inclusion to pay-per-use, and whether Anthropic discloses more detail on the hidden AI-research filters in response to community pushback.
Analytical support, not advice — assumptions and open questions stated above.